(18d) Model-Based Design of Catalysts

This team of researchers is working on a model-based approach to catalyst design called Discovery Informatics.  In this concept, the first objective is to identify descriptors that that can be linked to the kinetic performance of a family of catalysts through a quantitative forward model.  This demanding task requires a number of computational tools.  Since a microkinetic description is essential for the forward model, we have focused on tools for model discrimination and accurate extraction of rate constants from kinetic data.  For single site olefin polymerization, which is our homogeneous catalysis test bed, we have developed a parallelized population balance modeling tool which allows molecular weight distributions from gel permeation chromatography to be included in a multi-response data set that is used to verify mechanism and provide rate constants for each of the necessary elementary steps.  For the water gas shift (WGS) reaction on supported metals, our heterogeneous catalysis test bed, we have included Bayesian statistical analysis that provides specific accounting for error and Markov Chain Monte Carlo analysis to give probability distributions for parameters and their correlations with each other.  While descriptors to relate to rate constants can sometimes be identified through linear correlations from statistically designed experiments, we have turned in this developmental stage to density functional theory to help identify descriptors, to better define reaction pathways, and, particularly in the case of heterogeneous systems, to compute energies that are difficult to measure.

            Progress in building forward models will be discussed in terms of hexene polymerization over non-Cp coordination complexes of Zr and Hf and WGS over supported Au and Pt. In the homogeneous polymerization system we have discovered that the pendant ligand, NMe₂ versus THF, has a large effect on rate constant for chain transfer, while the change from Zr to Hf, lowers the propagation rate constant by a factor of 20.  We will interpret this data in light of DFT results and our previous finding that the ion pair separation energy is an important descriptor[i].  For the WGS catalysts, we have identified the corner and perimeter atoms in contact with the support as the active sites for supported Au, all surface atoms as active for supported Pt, and, in spite of these differences, that the support plays an important role for activation of water for both Au and Pt catalysts.  Models integrating these findings will be presented.

            The value of a successful forward model is that it can be used in a guided stochastic search of the descriptor space to identify optimum catalyst candidates.  Tools and information needed and obstacles which must be overcome to close this inverse cycle will also be discussed.